A convex formulation for multiple ordinal output classification
作者：
Highlights：
• First study multiple ordinal output classification (MOOC) as a general machine learning task.
• Propose an effective formulation to jointly model the relationship among multiple ordinal dinal variables of MOOC and their discrete ordinal values.
• Exemplify a convex objective function by the formulation which allows us to learn the optimal model parameters and the relationships among output variables simultaneously.
• Apply the kernel trick to provide a nonlinear extension to enhance nonlinear ability of our model.
• Demonstrate that our method not only achieves effective classification performance but also reveals the structures among output variables.
摘要
•First study multiple ordinal output classification (MOOC) as a general machine learning task.•Propose an effective formulation to jointly model the relationship among multiple ordinal dinal variables of MOOC and their discrete ordinal values.•Exemplify a convex objective function by the formulation which allows us to learn the optimal model parameters and the relationships among output variables simultaneously.•Apply the kernel trick to provide a nonlinear extension to enhance nonlinear ability of our model.•Demonstrate that our method not only achieves effective classification performance but also reveals the structures among output variables.
论文关键词：Multiple ordinal output classification,Multiple discrete ordinal variables,Ordinal regression,Relationships,Convex function
论文评审过程：Received 2 October 2017, Revised 10 August 2018, Accepted 6 September 2018, Available online 7 September 2018, Version of Record 15 September 2018.
论文官网地址：https://doi.org/10.1016/j.patcog.2018.09.005
